Run ComfyUI Remotely

Run ComfyUI on a remote GPU with full custom node support.

Steps

1. Create a pod

swm pod create -p runpod -g "RTX 4090" -n comfy \
  --lifecycle auto-down --idle-timeout 20 -y

2. Install ComfyUI

swm setup install comfyui <provider>:<pod-id>

3. Install custom nodes

Two paths — pick whichever fits the workflow.

Option A — ComfyUI-Manager (interactive). ComfyUI-Manager is preinstalled at /workspace/ComfyUI/custom_nodes/ComfyUI-Manager. After starting ComfyUI (step 4), open the UI and use the Manager panel to search for and install nodes. Good for exploration and one-off pods.

Option B — Scripted install (reproducible, agent-friendly). Install nodes before the first start so registration happens cleanly:

swm run <provider>:<pod-id> \
  "cd /workspace/ComfyUI/custom_nodes && \
   git clone --depth 1 https://github.com/example/my-node.git && \
   /workspace/ComfyUI/venv/bin/pip install -r my-node/requirements.txt"

Pin a specific commit with git checkout <sha> if you want bit-for-bit reproducibility across pods. This is the recommended path for agent-driven setups and CI.

swm does not bundle a curated extension list — the ComfyUI custom node ecosystem moves too quickly for that to stay useful, and Manager already solves discovery.

4. Start ComfyUI

swm setup start comfyui <provider>:<pod-id>

5. Access

ComfyUI will be available at the proxy URL or via SSH tunnel at http://localhost:8188.

6. Download workflows

swm download <provider>:<pod-id> /workspace/ComfyUI/output/ -d ~/Downloads